[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On 16/09/14 09:47, Tremayne, Steve wrote: >>> have you tuned your kernel? > > Well, pfSense is a fairly locked down BSD-based distro â so my answer is > ânoâ, and a longer answer is âer, Iâm not sure if I canâ I think he means TCP parameters, but TCP untuned will do more that 3MB over short WAN links. I used to get ~25MB/s over the Internet between two servers in the UK with gigabit interfaces. So unless the firewall is crazy slow, this is unlikely. > Cables are a mix of Cat5e and Cat6 in an accurate representation of my > intestine combined with my MP3 player headphone leads... and yep, some > run past and are probably wrapped around a domestic 50Hz white-noise > signal injector :o) Ditto cable. They could be at fault, but not likely. Interface speeds are irrelevant, it is the speed they are running at. Mis-negotiating is common but 3MB/s is neither 10Mbps or 100Mbps. However always worth checking the current settings. Amazing how many use to negotiate sub-optimal settings, things are better these days but I always check when troubleshooting network speed issues. Any time someone says rsync is slow (or NFS, or CIFs or anything complex) the first step is use a protocol which is fast and simple, piles all the data into a TCP stream as fast as it can. Old fashioned "ftp" (if security can be constrained) is a good choice. Or Google using netcat for the same sort of thing. Shuffle one big file that'll fit in buffer cache, send it twice, to eliminate read performance on the end reading. This should give you an indication of end to end network speed, now you know if it is rsync, or the network. The same approach works for measuring improvements in network speed. Traffic shaping (protocol specific) can of course fool this approach, but hopefully you know if traffic is shaped. There are some tools around that do same as ftp, but really VSFTP with an anon account on some copies of your kernel does it fine, and most ftp clients give a throughput figure (some need poking to do so). Other common issue is write performance sucking due to sync issues with data or metadata. bonnie++ will tell you if your IO sucks. But also iotop and sar should show you if it is doing insane IO behaviour during the rsync. -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq